IADR Abstract Archives
Machine-Learning of Volatile Organic Compound Patterns for Peri-Implant Disease Diagnosis
Objectives: Early detection of disease activity surrounding implants prevents irreversible damage for more effective treatment strategies. Diagnosis of peri-implant diseases (PID) currently relies on traditional periodontal assessments methods (radiographs, bleeding indices, probing indices), yet surfacing technologies potentially extend a more effective approach. Volatile organic compounds (VOCs) are released from microbial and host cells during cellular metabolic processes and biochemical alterations caused by disease. VOC inputs can be manipulated through algorithms for machine learning to enable disease differentiation and potential diagnosis. We aimed to: 1.Identify PID specific VOC patterns 2.Use sensor-array inputs to employ machine-based learning for the identification of PID.
Methods: During routine dental check-ups at the student’s dental clinic (The Hebrew University), 90 healthy patients (>18 y.o) were sampled and divided into 3 groups: 1. periodontally healthy (non-implant group) 2. patients with healthy implants 3. patients with peri-implantitis. Exhaled breath (EB) samples were. collected and PID-specific VOC patterns were identified using Gas Chromatography-Mass Spectrometry (GC-MS). Furthermore, 40-cross-reactive sensors (Electronic Nose (E-Nose)) were employed to distinguish EBs responses of the sensor-array via machine learning algorithms to identify PID.
Results: Presence of distinctive VOCs from 78 EB samples were associated with PID, differing by molecule types and relative abundance. Discriminant Functional Analysis plots of the nanosensor cross-reactive response to the collective VOCs enabled disease separation. Interestingly, presence of an implant (groups 2 and 3) showed statistically significant variation to the non-implant group.
Conclusions: VOC identification cross-referenced with the current gold standard of clinical diagnosis could serve to enhance machine learning algorithms for better diagnostic accuracy necessary for clinical application. This could provide a platform for early diagnosis and monitoring of PID to prevent acceleration of the pathology, reduce chair-side time, and contribute to the control of this chronic disease to enhance quality of life and health of patients.
Methods: During routine dental check-ups at the student’s dental clinic (The Hebrew University), 90 healthy patients (>18 y.o) were sampled and divided into 3 groups: 1. periodontally healthy (non-implant group) 2. patients with healthy implants 3. patients with peri-implantitis. Exhaled breath (EB) samples were. collected and PID-specific VOC patterns were identified using Gas Chromatography-Mass Spectrometry (GC-MS). Furthermore, 40-cross-reactive sensors (Electronic Nose (E-Nose)) were employed to distinguish EBs responses of the sensor-array via machine learning algorithms to identify PID.
Results: Presence of distinctive VOCs from 78 EB samples were associated with PID, differing by molecule types and relative abundance. Discriminant Functional Analysis plots of the nanosensor cross-reactive response to the collective VOCs enabled disease separation. Interestingly, presence of an implant (groups 2 and 3) showed statistically significant variation to the non-implant group.
Conclusions: VOC identification cross-referenced with the current gold standard of clinical diagnosis could serve to enhance machine learning algorithms for better diagnostic accuracy necessary for clinical application. This could provide a platform for early diagnosis and monitoring of PID to prevent acceleration of the pathology, reduce chair-side time, and contribute to the control of this chronic disease to enhance quality of life and health of patients.